Google launched its advanced AI reasoning model Gemini 2.5 on Tuesday with a design that improves problem-solving by incorporating a “thinking” pause prior to response generation. Google achieves a major milestone in AI advancement by incorporating advanced reasoning features into its models to boost both accuracy and contextual comprehension.

Google launched its new AI model family with the unveiling of Gemini 2.5 Pro Experimental. The company claims this multimodal AI model represents its most intelligent and powerful system developed so far. Developers and advanced users can access Gemini 2.5 Pro Experimental through Google AI Studio starting Tuesday. The company offers access to this service for users who subscribe to its $20 monthly Gemini Advanced plan through the Gemini app.

AI Reasoning Models Are Reshaping the Industry

The tech industry has shown significant interest in AI reasoning models after OpenAI established this domain with its initial reasoning model o1 in September 2024. The leading AI firms Anthropic, DeepSeek, xAI and now Google have entered competition to create and improve similar models since that time. Reasoning-based AI systems require more computing power and longer processing time to validate information and improve problem-solving performance before delivering an answer.

AI systems that incorporate reasoning techniques have shown significant progress especially within complex fields like mathematics and coding. The tech community widely believes reasoning models will prove essential for building autonomous AI agents which perform tasks without needing much human oversight. The high computational demands these models require result in greater operational costs.

Google’s Efforts to Improve AI Reasoning

Google has previously explored reasoning-based AI before this current initiative. A preliminary version of Gemini featuring initial “thinking” capabilities was unveiled by the company in December 2024. Gemini 2.5 stands as Google’s most advanced and competitive model designed to exceed OpenAI’s acclaimed “o” series models.

Google demonstrates that the Gemini 2.5 Pro model surpasses both their earlier advanced AI models and several top competitor models in different testing benchmarks. Gemini 2.5 stands out as Google’s solution for creating visually stunning web applications and coding-heavy agent applications.

Benchmark Performance: How Gemini 2.5 Pro Compares

Gemini 2.5 Pro earned 68.6% on the Aider Polyglot benchmark meant to measure code-editing proficiency. Gemini 2.5 Pro outperforms leading AI models from OpenAI, Anthropic and the Chinese AI lab DeepSeek.

Gemini 2.5 Pro registered a score of 63.8% on the SWE-bench Verified benchmark which evaluates general software development capabilities. The Gemini 2.5 Pro model ranks higher than both OpenAI’s o3-mini and DeepSeek’s R1 but remains below Anthropic’s Claude 3.7 Sonnet which scored 70.3%.

The multimodal evaluation Humanity’s Last Exam which consisted of thousands of crowdsourced questions across mathematics, humanities and natural sciences found Gemini 2.5 Pro achieving a score of 18.8%. Gemini 2.5 Pro ranks above the majority of flagship models currently available in its sector.

Expanding Context Window for Enhanced Processing

The expanded context window stands out as a key feature of Gemini 2.5 Pro. The current version of the model supports a maximum input size of 1 million tokens which corresponds to about 750,000 words. The expanded context window in Gemini 2.5 Pro surpasses the total length of the “Lord of the Rings” book collection. Google plans to expand its processing capacity which will soon allow for 2 million token input lengths.

What’s Next for Gemini 2.5 Pro?

Google has yet to announce what the API pricing for Gemini 2.5 Pro will be. The company stated that additional information about pricing and commercial availability will be shared in the upcoming weeks.

Google’s release of Gemini 2.5 proves its ongoing dedication to developing AI reasoning solutions. The company’s recent advancements suggest that AI models will grow more sophisticated and intelligent while becoming better equipped to solve complex problems with higher accuracy as competition in the field grows stronger.